A Method of Entire Shape Recovery of a Solid by Surrounding Cameras

ステレオ法を用いた立体の3次元形状復元では,カメラで共通に観察される部分しか復元されないため,全周モデルを作成するには部分的に復元された形状をつなぎ合せる必要がある.このため厳密なカメラキャリブレーションが必要であり,3次元復元の効率化を妨げている.本稿では,因子分解法の適用により,簡便なカメラキャリブレーションに基づく,つなぎ合せを行わない全周復元法を提案する.まず物体を囲むように複数のカメラを配置し,全カメラで共通に観察される物体上の特徴点に因子分解法を適用して,全カメラの方向および特徴点の3次元座標を求める.次に前面カメラの方向および前面カメラでしか観察されない物体上の特徴点の画像座標から,それらの特徴点の3次元座標を計算し,また後面カメラの方向および後面カメラでしか観察されない物体上の特徴点の画像座標から,それらの特徴点の3次元座標を計算する.この手続きにより,物体の全周の特徴点が復元される.実験により提案法の有効性が示された.The existent 3-D modeling technique such as the stereoscopic vision recovers 3-D shape of part of a solid which can be observed commonly from the set cameras. This inevitably needs registration among recovered partial shapes in order to obtain an entire 3-D model of the solid For this purpose, strict camera calibration is indispensable, resulting in not very efficient modeling procedure. In this paper, a technique based on a simple camera calibration is proposed for recovering entire shape of a solid without registration by the application of the factorization. In the first place, multiple cameras are set around an object interested and the feature points observable from all the cameras recover their 3-D locations by the factorization. By employing the information on the orientations of the cameras observing the front part of the object, those feature points observable only from the frontal cameras recover their 3-D locations. In the same way, the feature points observable only from the cameras looking at the rear part of the object recover their 3-D positions by the employment of the orientations of the rear cameras. In this way, all the feature points on the object recover the 3-D positions, providing entire shape of the object. Experimental results show satisfactory performance of the proposed technique